A Teacher Behavior Recognition Model Based on Multi-Stream Graph Convolution Network
Xiaoyu Ma, Juxiang Zhou, Di Wu, Sen Luo · 2023
In recent years, various innovative applications of intelligent education driven by the new generation of AI technology have become research interest in the fields of education science and technology. Recognition of classroom teaching behavior based on image and video analyses has attracted much attention of researchers in recent years. Compared with the research on classroom student behavior, the research on classroom teacher behavior recognition and analysis is still in its infancy. Recognition of teacher behavior are associated with issues such as the lack of teacher behavior data sets in real classroom scenarios, difficulty in migration of existing behavior identification algorithms, and low accuracy of teachers' classroom behavior identification. Accordingly, this paper constructs a data set of teacher behavior in the real classroom environment, including video and bone information data of five types of teacher behavior, and proposes a teacher behavior recognition model based on multi-stream graph convolution network. This model effectively fuses joint, bone and motion information; adds graph attention module on the basis of graph convolution; learns and optimizes the connection between nodes; and finally realizes effective recognition of teacher behavior in classroom videos. The proposed model was compared with different models on self-built datasets for several groups of experiments. Our results show that the proposed model performs well in teacher behavior recognition tasks in real classroom scenarios.